Mechwarrior: Dissecting a Contender

Once things got moving with BF’s Mech-Breeding Genetic Algorithim the fur really started to fly. Spurred on by random memories framed in spontaneous theorizing and questioning, BF narrowed down the fitness function to one which did not fly in the face of what we consider to be reason, and began to produce fairly fit fighters fitting fighting factors we felt were fundamental.

Refined Fitness Function Parameters:
  • Range was ‘negotiated’ between the mechs with each seeking to strike from optimal distance. Both calculated the optimal range where they would do more damage than their opponent. If their optimal range was farther back, they would walk backward or jump full speed, if it was closer they would run forward full speed. Non-jumping movements lost 1-3 random movement squares due to terrain.
  • ‘Mechs were enabled to breed, taking the chassis from one and randomly adding components back from each until internal space and weight limit was reached.
  • Two Gunnery levels were tested: Gunnery 3 and Gunnery 1
How we got there:

At the end of a run, the 20 top mechs were put through a survival test / sanity check to see how they fared against 2 pre-determined opponents: a standard atlas, and 55 tonner, moving 5-8-5 with a single ac2. These factors and the final concept checking battle were reached primarily because of the following observations:

BF previously commented on the interim results:

I’ve been doing multiple runs while varying different conditions and the things that show up as important are very telling. For example, gunnery skill. In a competition with Medium mechs only, when I run with gunnery skill of 4, I get mechs with 1 large laser and a few mediums with 8 run speed. One run with these stats practically duplicated the phoenix hawk.
Lower the gunnery to 2 or 0, and the optimal mech has a run of 3! with LRM5’s a PPC and 5+ medium lasers and heat sinks to fire them. There is no need to move if you can hit from any range, and no advantage to moving if your opponent can still hit you.
He later followed up with:
The assault results are in as well with the same pattern. 4 gunnery is 90 ton with a 5 run, 3 large lasers, and a mix of small and medium, maybe 1 lrm5 pack. 0 gunnery is a 100 ton, 1 walk weapon platform with LRM 20, a couple of 10s, ac10 and medium lasers.
I tossed in:
In my memories of play, with only gunnery and piloting being considered, by far the more often-used skill was gunnery. Still… a failed piloting roll has some very serious consequences, and determines the possibility of physical attacks…
Which BF followed up with:
I think my next steps are to assign ratings to the weapons similar to the plus/minus system they have to rate players in hockey. I can also have some of the standard mechs face off in duels. Any standard mechs you want to test out? Wolverine vs Warhammer perhaps? 🙂
Douglas added:
I hadn’t even considered the pilots, but I see your point. It does seem to bring out an interesting one-sidedness to the game mechanics.As Runeslinger pointed out, Gunnery was always more important than piloting. What I was just thinking about now is that the game, while reflecting a pilot’s ability to shoot well, does not reflect a pilot’s ability to manuver or pilot well in combat (only whether or not he can keep the mech from falling over). 

The whole concept of the mechs (at least the mediums and lights) is that they replaced tanks because of their manuverability and versatility. On top of that, the big mechs are supposed to be slower. In the “fiction” they are deemed too valuable and specialized to be deployed by themselves—without support troops and/or light and medium mechs.

The game treats all pilots the same for defence. It only factors speed and terrain, as if all pilots would handle this or more less the same. I hadn’t thought too much about how this element is missing altogether.

The computer games, however, generally reflect this to some degree. When you are piloting a big mech, you have a hard time when going up against a skilled player running around like a madman in a little light mech. Even if you are a good shooter and a good player, too, he’s hard to hit because your mech is so slow compared to his, but he’s also manuvering well and throwing you off. Your big mech is much better suited to taking out other big mechs, turrets, and Hell, entire towns.

A bad player, on the other hand, you will still hit easily no matter how fast he’s running or what terrain he’s in, because he sucks and you are a good shot. Here it starts to resemble the board game.

Yeah, in the board game once you lower gunnery scores enough everything becomes irrelevant and you might as well just sit still in this biggest gun platform you can get. I think what you’ve found is quite accurate.

BF responded with:

They’ve already factored tactics into initiative which gives that skill a huge boost.

Just out of curiosity, I’ve set the 55 tonners against each other and I think you’ll be surprised at the winner. The scorpion was the bottom of the list having never won a match. No one will be surprised there. Griffin was second last, which surprised me, but I suppose a sniper isn’t that great one on one. Shadowhawk and Wolverine would switch back and forth with number of wins, but it was actually the Shadowhawk that inched out the Wolverine in the end. The surprise winner was the Dervish. Completely undefeated against all the others in every simulation I ran. All of the mechs were stock, with no modifications.
By the way, the Warhammer fares really badly in my tests, overheating badly when it fights nearly any mech. Even the 500 year old Hammerhands takes it out every single time.

My response was limited to:

The secret is in the sauce

Douglas, inspired to speculate by these teaser results commented:

If the ranges are random, I can see the Shadowhawk winning in the medium class. It has a nice balance of weapons for all ranges, with no real strength in any (if I remember correctly).

I was thinking again about how speed and tactics worked well when the gunnery scores were at normal levels. The reason why mechs with a focus on medium range weapons could rarely even hit me, is that if I allowed a shot at all I’d still feel very safe at 9 hexes. At maximum range for their medium lasers, and mine for my medium range weapons, my odds were still better due to speed and terrain tactics giving me more defence. I’d usually hit with some of my weapons and usually be missed altogether. Even better was a PPC variant! With superior mobility and range, I would never allow a shot at all. I could always be at the range I wanted and punch holes in them from outside their maximum range.

Is there anything in your coding that allows for the faster mechs to use their speed to be in the best possible range for both defence and offence, while the slower mech is forced to take what shots they can get? Would mechs instead choose to kill another mech slowly if they could do it at little or no risk to themselves? Adapting to the mech they are faced with…? Or would they just be at random ranges and unload with what they have available…?

This is one of the main reasons why, when combined with Runeslinger’s tactics, his Warhammer was almost always a threat. Like a few other mechs, you always had to worry about those twin PPCs if you were a medium-mech player trying to skirt his ranges. They were arm mounted as well, so combined with a torso twist, it made it almost impossible to hit him on his “blind” side. It had something for every range. Most medium mechs, other than some AC2 variant freak (I made one once), will have to get into his PPC medium or short ranges, before they can really make effective attacks themselves. Then he has “back up” weapons and more armour to help out.

It was not at the optimum tonnage for its movement according to my math and comparisons, but it’s look, concept, and overall style had to count for something. It even had a light that could get shot off. 🙂

So what worked for the Warhammer is that it didn’t need to chase. I’d have to approach one very carefully. I could easily take on any slow mech with a lot of medium lasers but that twin PPC punch had a long reach. The tactical advantage made the PPC awesome.

Ton per damage-per-second, the medium laser is the best weapon in the game. Always have one or two, I feel. However, combined with superior mobility and tactics it can be rendered mostly ineffective. Against a mech with a PPC and greater mobility, you’re pretty much screwed no matter how many medium lasers you have. It becomes like a bigger version of the small laser at that point.

Get in close on a slower target and hit it with a mountain of medium lasers and you’ll realize the theory in reality, I think. No sane player will let you do that if they can help it, though.

I think I should say that I didn’t like the Wolverine so much as I liked it’s movement for its weight class (maximum number of tons left over for armour and weapons for a mech walking 5, running 8, and jumping 5). I’d always play a variant if I could.

When the Uziel, the 50 or 55-tonner with twin PPCs, appeared I quickly forgot about the Wolverine. The new mech looked cool on top of it all. 🙂

My variants tended to favour:

(1) At least one PPC for long range punch… One is usually all that could be managed effectively. Anything that can’t match my range and mobility is pretty much dead with smart tactics. I’ll make it impossible for them to hit. Even one PPC is enough to ruin the day of any mech if it can’t shoot back.
– Sure it had a minimum effective range, but being that close was generally a bad idea anyways (where medium lasers and AC 20s can finally say hello to short range). Then they later put extended range PPCs in with no minimum effective range in the game… Get extra heat sinks and you’re good to go. There’s your main gun.

(2) At least one medium laser, preferably two… You will get into short/medium range or there will be mechs that can catch you from time to time, so never be without.
– Best ton per damage-per-second ratio among other things. Only mechs that can’t afford bigger weapons rely on these as primary offence. You have to close within 3 hexes to be in short range. Even small lasers can shoot at that range. Big mechs have a lot of big weapons—you can’t let them hit you with them often. For mechs that can catch you or for when you close for the kill, you’ll need these as backup weapons.

(3) Something else… An alternate strategy or trick to complement the PPC ‘standard’ and medium laser ‘backup’.
– One thing that worked really well for me was getting rid of the SRM-6 in favour of an SRM2. It was cheaper and left more tons for other things like heat sinks, came with more ammo, and was the only launcher in the game to come with inferno missiles. Better than a flamer for both overheating an opponent (in one shot sometimes—especially if they just unloaded and missed with everything) and for range, it was a cheap and effective weapon to shut an enemy down.

This was the thing that most players never saw coming. After dancing with them at farther ranges and them always pushing their heat levels, an opportunity comes up that says “Attack now!”. Close and unload with everything including this. Your mech should be designed to be able to do a full salvo for several turns—preferably endlessly. One hit from infernos and you cripple a mech that was already overheating some. They can either: (1) let you shoot them while they try to cool, or (2) overheat further and really risk shutting down.

On top of that you could set the map on fire and create smoke “cover” to complement your tactics and so on. Fun stuff…

Also, I know that many players swear that jump jets are useless, but I couldn’t imagine playing without maximum jump range. Even the fastest Locust would get cornered eventually, and there were so many more ‘you can’t hit me’ options for use of terrain with jumping. In fact, a fast-medium tactics game relies on them heavily, I think.

Sometimes you could even block a shot altogether by taking away line of sight. I forgot how many times I simply jumped behind a wall, building, or range of hills after leading other mechs into what they thought was my doom. I could always move 5 hexes in any direction to land in the terrain of my choice while other mechs lost their movement advantage or struggled even further in rough terrain or in cities.

BF ended the discussion with the final round-up of ideas:

Douglas, your instinct is right on the money, and tested with a few modifications to the algorithm mentioned above. I changed the random distance to a ‘negotiation’ between the two ‘mech’s speed. Each ‘mech tries to reach their optimal range decided by their own and their opponent’s damage potential. ‘Mechs whose optimal is closer than current run or jump forward, ‘mechs who’s range is farther away, walk or jump backwards. I added some random terrain to make jumping more valuable. It was as close to simulating your strategy as I could without programming a full game 🙂

Heavy/Assault Mechs still add 4 or 5 medium lasers but now they throw in a few AC5 and PPCs. Some runs show preferences for a single LRM 10.

4 Gunnery still tends to the laser side of things, lower tends to LRM, AC and SRM, though  still with plenty of lasers.

The ‘mechs that the algorithm is discovering are doing a great job at taking out ‘mechs from the book, like an Atlas, but I created the exact Wolverine with AC2 that you mentioned, and no-one has beaten it yet. 🙂

The results, when done, looked like this:
  • All ‘mechs, regardless of size, evolve toward Medium Lasers plus at least one long-range weapon, tons of heat sinks, and lots of armour.
  • The open weight class ends up 5-7 Medium Lasers, 2-4 long-range weapons, and a screaming running speed of 2.
  • Limiting to Medium or Light ‘mechs, we get 4-5 Medium Lasers, typical 5-8 movement (no jump, surprisingly), and one or two long-range weapons
  • Choice of long-range weapons depends entirely on Gunnery skill. With a Gunnery of 4, Large Lasers, and PPCs win. With Gunnery 2 or less, LRM10s and AC5s are ideal. Gunnery 2 or less will often add an SRM6 as well.

Who won? BF summarized the results with this telling quote:

It depends. 🙂

He followed up with a challenge:

You know? I take that back. There is a winner. One mech to rule them all.

Think back to your 3025 mechs, and which one epitomizes the results we saw here?

2 Responses to “Mechwarrior: Dissecting a Contender”
  1. Runeslinger says:

    Contrary to personal experience of certain ‘mechs like the Warhammer and Marauder being extremely deadly in actual play, this grand experiment placed their designs near the bottom of the rankings.

    Would anyone care to wager a guess as to which designs fared best under these conditions?

    I will post a full disclosure of the findings in a subsequent post.

Check out what others are saying...
  1. […] Runeslinger mentioned this post in a further discussion of evaluating mech design superiority[…]

Speak your piece~

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

  • Revelations of Glaaki

  • Invocation

    Do not summon up that which you cannot also put down:

    runescastshadows at the intersection of Google and Mail.

    Find us on Google+

  • Role-Playing Stack Exchange

%d bloggers like this: